Add DBConvert Streams (Federated SQL)

[slotix]

What this adds

dbconvert-streams — a read-only MCP server for PostgreSQL, MySQL, S3-compatible
buckets and folders of Parquet / CSV / JSON files. Its distinguishing feature is
that one query can join across those sources: a live PostgreSQL table against
a Parquet file in a bucket, or a CSV against the table it was exported from, with
no copying and no staging step. Federation runs in-process on DuckDB.

21 tools, all of them reads. There is no tool that writes — not disabled, absent
— and every tool carries readOnlyHint, with destructiveHint: false and
openWorldHint: true stated explicitly.

About the image

The image is our own (slotix/stream-mcp) rather than one built from source in
this PR: the server is proprietary software, distributed as a binary. It carries
LABEL io.modelcontextprotocol.server.name="com.dbconvert/streams", and the same
build is published in the official MCP registry as com.dbconvert/streams, where
the DNS-verified dbconvert.com domain vouches for it.

source.project points at our public repository, which holds the documentation,
examples and release artifacts for the product; the server binary itself is not
open source. If a Dockerfile in a public repository is a hard requirement, tell
us and we will find another shape.

Configuration

• connections — connection strings, repeated. postgres://…, mysql://… or
  s3://bucket/prefix?region=…, each optionally prefixed with name= to choose
  what the source is called in chat.
• paths — folders of data files, mounted and passed to the server, following
  the pattern used by filesystem.
• AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY — needed only for S3-compatible
  storage that is not AWS (MinIO, R2, Spaces). Keys are never read from the URL.

Verification

• task validate -- --name dbconvert-streams passes every check.
• task catalog -- dbconvert-streams generates the catalog entry.
• The tool list in tools.json was captured from the running container with the
  MCP inspector, not written by hand.
• End to end against live databases: docker run -i --rm slotix/stream-mcp "pagila=postgres://…" "sakila=mysql://…" "lake=s3://…" /data lists all four
  sources and answers a federated join across them.

Docs: https://streams.dbconvert.com/docs/mcp/standalone